Iam currently conducting a preliminary analysis, before the CFA and SEM analysis. I have a few negative items (likert-scala). Do I need to permanently reverse code those negative items before conducting a test of normality, means, skewness, kurtosis and SD? And also, must those negative items be reverse coded before I start with the CFA and SEM-analysis?
I appreciate the good help!
@Moin Yes you need to reverse code negative polarity items even before doing data screening/prepration (checking for missing, unengaged responses, outliers) as all the items need to be on the same polarity.
Moin Deljo It depends. If you use the items directly as indicators of latent variables in CFA/SEM, then recoding may not be necessary. The negative items would simply have loadings that are opposite sign of the positive items. That being said, negative loadings (although perfectly appropriate and interpretable) may be difficult to communicate in a research report. Also, using items of different polarity can sometimes cause convergence problems in CFA/SEM because of the default starting values in software programs. Those convergence issues can typically be easily resolved by providing appropriate starting values for the factor loadings (i.e., positive values for positive items and negative starting values for negative items) so that this isn't a really serious issue.
Recoding also does not change the item distribution shape, so that it does not matter whether you examine item distribution shape before or after recoding.
On the other hand, when you analyze scale (composite, sum) scores or item parcels that are a function of multiple items some of which are positive and some of which are negative, then you definitely need to recode either the positive or the negative items before computing such scale scores or parcels. The reason is that a scale score only makes sense when all components are positively correlated. After recoding, all items should be positively correlated.
